What problem does it solve? Applications calling LLM providers have no visibility into what each request costs, how many tokens it uses, or how long it takes. This Skill integrates the Caveman gateway into an existing repository so every LLM request is measured without changing application behavior. ## Core Features & Use Cases - Callsite Discovery: Scans dependency files and source code for OpenAI, Anthropic, Vercel AI SDK, LangChain, LiteLLM, and raw HTTP LLM callsites before making any changes. - Gateway Wiring: Rewrites each callsite's base URL to the gateway path with an app slug and adds the x-cave-api-key auth header, supporting both stored and byok provider-key modes. - Verified Integration: Sends one real verification request through the new wiring and reports the actual HTTP status and token usage, with verbatim failure templates for unreachable gateways, invalid keys, and routing errors. - Use Case: A developer pastes the Caveman setup prompt into their support-bot repo; the Skill finds the OpenAI SDK client, points it at the gateway, verifies with a real request, and the spend appears on the dashboard grouped under the app slug. ## Quick Start Paste the Caveman setup prompt containing the gateway URL and Cave API key, or say "set up caveman" to wire this repository's LLM calls through the gateway.
